Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Credit risk data cleaning and variable screening pipeline for pre-loan modeling. Use when working with raw credit data that needs quality assessment, missing value analysis, or variable selection before modeling. it covers data loading and formatting, abnormal period filtering, missing rate calculation, high-missing variable removal,low-IV variable filtering, high-PSI variable removal, Null Importance denoising, high-correlation variable removal, and cleaning report generation. Applicable scena
.claude/skills/datanalysis-credit-risk/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | — | — |
| case-03 | ✗→✓ | ▲ Improved | — | — |
| case-07 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-01 | ✗→✓ | ▲ Improved | — | — |
bash# Run the complete data cleaning pipeline python ".github/skills/datanalysis-credit-risk/scripts/example.py"
The data cleaning pipeline consists of the following 11 steps, each executed independently without deleting the original data:
| Function | Purpose | Module | |------|------|----------| | get_dataset() | Load and format data | references.func | | org_analysis() | Organization sample analysis | references.func | | missing_check() | Calculate missing rate | references.func | | drop_abnormal_ym() | Filter abnormal months | references.analysis | | drop_highmiss_features() | Drop high missing rate features | references.analysis | | drop_lowiv_features() | Drop low IV features | references.analysis | | drop_highpsi_features() | Drop high PSI features | references.analysis | | drop_highnoise_features() | Null Importance denoising | references.analysis | | drop_highcorr_features() | Drop high correlation features | references.analysis | | iv_distribution_by_org() | IV distribution statistics | references.analysis | | psi_distribution_by_org() | PSI distribution statistics | references.analysis | | value_ratio_distribution_by_org() | Value ratio distribution statistics | references.analysis | | export_cleaning_report() | Export cleaning report | references.analysis |
DATA_PATH: Data file path (best are parquet format)DATE_COL: Date column nameY_COL: Label column nameORG_COL: Organization column nameKEY_COLS: Primary key column name listOOS_ORGS: Out-of-sample organization listmin_ym_bad_sample: Minimum bad sample count per month (default 10)min_ym_sample: Minimum total sample count per month (default 500)missing_ratio: Overall missing rate threshold (default 0.6)overall_iv_threshold: Overall IV threshold (default 0.1)org_iv_threshold: Single organization IV threshold (default 0.1)max_org_threshold: Maximum tolerated low IV organization count (default 2)psi_threshold: PSI threshold (default 0.1)max_months_ratio: Maximum unstable month ratio (default 1/3)max_orgs: Maximum unstable organization count (default 6)n_estimators: Number of trees (default 100)max_depth: Maximum tree depth (default 5)gain_threshold: Gain difference threshold (default 50)max_corr: Correlation threshold (default 0.9)top_n_keep: Keep top N features by original gain ranking (default 20)The generated Excel report contains the following sheets:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-23 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +78 percentage points is the difference between those two pass rates over the 22 comparable cases.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.